ETO AGORA FY2026 NDAA, Section 6603 ("Application of artificial intelligence policies of the intelligence community to publicly available models hosted in classified environments")

FY2026 NDAA, Section 6603 ("Application of artificial intelligence policies of the intelligence community to publicly available models hosted in classified environments")

Proposed 2025-03-14 | Enacted 2025-12-18 | Official source

Summary

Ensure the Director applies AI policies to publicly available models in classified environments. Require the Chief AI Officer to establish AI testing standards considering risk and ensure secure environments for model evaluation. Prevent intelligence community authority over model alterations for viewpoint bias.

  • This summary is awaiting validation (peer review by a second AGORA editor).

Key facts

🏛️ This document has been enacted by the United States Congress. For authoritative text and metadata, visit the official source.

🎯 This document primarily applies to the government, rather than the private sector.

📜 This document's name is National Defense Authorization Act for Fiscal Year 2026, Section 6603 ("Application of artificial intelligence policies of the intelligence community to publicly available models hosted in classified environments"). AGORA also tracks this document under the name FY2026 NDAA, Section 6603 ("Application of artificial intelligence policies of the intelligence community to publicly available models hosted in classified environments"). It is part of FY2026 NDAA.

↳ This document is part of a longer one: FY2026 NDAA. Some AGORA documents are "split off" from longer documents that mix AI and non-AI content, such as omnibus authorization or appropriations laws in the United States Congress. Read more >>

Themes AI risks, applications, governance strategies, and other themes addressed in AGORA documents.
  • Thematic tags for this document are awaiting validation (peer review by a second AGORA editor).

Risk factors (6)

Governance strategies (5)

AI application areas (2)

Full text

  • This is an unofficial copy. The document has been archived and reformatted in plaintext for AGORA. Footnotes, tables, and similar material may be omitted. For the official text, visit the original source.
  • Thematic tags for this document are awaiting validation (peer review by a second AGORA editor).
SEC. 6603. APPLICATION OF ARTIFICIAL INTELLIGENCE POLICIES OF THE INTELLIGENCE COMMUNITY TO PUBLICLY AVAILABLE MODELS HOSTED IN CLASSIFIED ENVIRONMENTS. (a) In General.--Section 6702 of the Intelligence Authorization Act for Fiscal Year 2023 (50 U.S.C. 3334m), as amended by section 6602 of this Act, is further amended-- (1) in subsection (a), by striking ``subsection (c)'' and inserting ``subsection (e)''; (2) by redesignating subsection (c) as subsection (e); and (3) by inserting after subsection (b) the following: ``(c) Application of Policies to Publicly Available Models Hosted in Classified Environments.--In carrying out subsections (a) and (b), the Director shall ensure that the policies established under such subsections apply to the greatest extent possible to artificial intelligence models generally available to the public and hosted in classified environments.

Requires the Director to apply AI policies of the intelligence community to publicly available models hosted in classified environments.

Requires the Director to apply AI policies of the intelligence community to publicly available models hosted in classified environments.
``(d) Testing Standards and Benchmarks.-- ``(1) Establishment.--The Chief Artificial Intelligence Officer of the Intelligence Community, or any officer designated by the Director of National Intelligence, shall establish standards for testing of artificial intelligence models in proportion to risk, including benchmarks and methodologies for the performance, efficacy, safety, fairness, transparency, accountability, appropriateness, lawfulness, and trustworthiness of artificial intelligence models across common use cases, including machine translation, object detection, and object recognition. ``(2) Identification of computing model.--The Chief Artificial Intelligence Officer of the Intelligence Community, in coordination with the Chief Artificial Intelligence Officers of the elements of the intelligence community, shall set standards for an appropriate secure computing environment, at a level (or multiple levels) of classification deemed appropriate, for elements of the intelligence community to engage in testing and evaluation of models prior to acquisition.''.

Requires the Chief AI Officer to establish AI model testing standards for risk, performance benchmarks, efficacy, safety, fairness, transparency, accountability, trustworthiness across ML, object detection and recognition use-cases; requires the Officer to set standards for a secure computing environment for model testing and evaluation prior to acquisition.

Requires the Chief AI Officer to establish AI model testing standards for risk, performance benchmarks, efficacy, safety, fairness, transparency, accountability, trustworthiness across ML, object detection and recognition use-cases; requires the Officer to set standards for a secure computing environment for model testing and evaluation prior to acquisition.
(b) Rule of Construction.--Nothing in subsection (c) or (d) of section 6702 of the Intelligence Authorization Act for Fiscal Year 2023 (50 U.S.C. 3334m), as added by subsection (a)(3) of this section, shall be construed-- (1) to authorize an officer or employee of the intelligence community to direct a vendor or prospective vendor to alter a model to favor a particular viewpoint; or (2) to apply to, limit, restrict, or otherwise affect any department, agency, or division or function of a department or agency outside of the intelligence community. (c) Updates.--The Director shall make such revisions to the policies issued under subsections (a) and (b) of section 6702 of such Act as the Director considers necessary.

Prohibits intelligence officers from directing vendors to alter AI models to favor specific viewpoints, and clarifies the restriction applies only within the intelligence community.

Prohibits intelligence officers from directing vendors to alter AI models to favor specific viewpoints, and clarifies the restriction applies only within the intelligence community.